Assembler Optimizer

(sideline comment: I totally approve of pyramidwarpex btw )

There are indeed a lot of different syntax conventions! and some of them incompatible. For example, I see that jasm uses "." to indicate local labels, were as asMSX uses "@@" for this, and uses "." for built-in functions (e.g., to print debug messages during compilation). So, if I want to support all of them, I'll need some command-line flags to indicate the parser to use...

But it's ok, this was very helpful, as I've been looking ant each assembler supports A LOT of specialized syntax, but most of those are rarely used. So, I just wanted to see which are the common constructs that are actually used in MSX projects, to support at least most of those. From what I see, if I add support for local labels, and a few more macros, I might cover already 50% of the projects linked above, so, I will be satisfied with that

And yeah, an option to output plain Z80 assembly without macros, etc. would be a great feature for ALL assemblers. In that way, we would be able to share code across assemblers, while still being able to code with our favorite one!

Yeah, assembler syntax is tricky. I see that with their best intentions in mind, each assembler developer expanded on the plain Z80 notation to make programming easier (although some of the decisions in a few are questionable, but that's a different discussion hehe). But every assembler went in a different direction, and now they are all mostly incompatible except for the most basic core features.

In any case i have a basic parser done (which I might expand bit by bit to support more syntax) and this weekend I'll get started with generating a dependency graph given a block of assembler code. Detecting whether an instruction depends on another is easy when they interact via a register or flag, but it gets tricky when they interact via a input/output port or via memory (unless the memory address can be resolved at compile time). So, I'm planning on being conservative and err on the side of adding more dependencies than maybe needed. The result will be that maybe I will miss some possible optimizations, but at least all optimizations that are applied will be guaranteed to not break the program (again all assuming non-self modifiable code).

I think both assemblers wil fail, because they are incapleble to handle these labels. (Yes i've tried)

Just checked with GuyveR800 (the creator of tniASM):
- In tniASM, right now you can define those Dos error codes or any labels starting with a period by using macros, For example %macro .NCOMP \ 0FFh \ %endmacro
- tniASM already supports labels that do not participate in the parent/local mechanism
- Support for such labels starting with a period was already planned for the next version. So .NCOMP equ 0FFh will work as well